Willingness-to-pay space and preference space: a unified treatment of equivalence, heterogeneity and interpretation
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "Willingness-to-pay space and preference space".
Warren: Here's the gist of the paper. A growing number of choice modelling studies estimate their models “in willingness-to-pay (WTP) space”, yet often treat this as a different model that fits better (or worse) than “preference space”, and misread what its parameters represent WTP space is a reparameterisation, and with non-random coefficients the two spaces give identical likelihoods, WTP estimates and standard errors Willingness-to-pay space as a reparameterisation The paper introduces WTP space as a reparameterisation of utility where the marginal utilities of non-cost attributes are written as the product of a cost coefficient and a monetary valuation, rather than “in preference space” This transformation involves changing the numeraire of utility rather than the units of an attribute For instance, writing Vjn = δj + βtc tcjn + βtcβv,tt tt jn + βtcβv,ch chjn (3) confirms that the WTP is positive for an attribute that reduces utility The paper explains how this reparameterisation is equivalent to a change of units if time entered the utility function in hours rather than minutes Equivalence and equivalence conditions Principle 1 states that a preference-space and a WTP-space specification with the same specification of the error term are the same model if and only if, up to the normalisations needed for identification, they imply for every person the same alternative specific constants and the same partial derivatives of utility with respect to every attribute, measured as in the data With random heterogeneity, the condition is that the joint distribution of these constants and partial derivatives across people is the same Proposition 1 confirms this equivalence by stating that if one model has maximum likelihood estimates obtained by estimating each model separately, then (i) the maximised log-likelihoods are identical and (ii) the asymptotic covariance matrix of ϕˆ obtained from WTP-space estimation equals the Delta method covariance J Var d(ˆθ) J Deterministically heterogeneous models When deterministic heterogeneity is present, the WTP measure is always defined as the ratio of partial derivatives in (2) The paper details how heterogeneity can enter in two places: through monetised utility via functions qtc,n or qk,n, or through the cost coefficient βtc,n For example, when income affects cost sensitivity with elasticity λρn = (incn/inc)λ, the correct WTP-space counterpart puts the interaction on cost within monetised utility as in equation (12) Non-linearity and practical limitations The convenience of WTP space, namely that the estimated parameters are the WTP measures, holds only when cost and all attributes enter monetised utility linearly With non-linear price effects, WTP must again be computed as a ratio of partial derivatives with Delta method standard errors, exactly as in preference space The paper suggests retaining a single directly estimated value at a reference point by centring covariates or normalising non-linear transformations at a reference point Random heterogeneity and distribution choices With random heterogeneity, the two spaces are no longer automatically equivalent, and differences in fit come from the distributions specified The equivalence of latent class models holds if each class has a non-random cost coefficient and the same class allocation probabilities However, discrete mixtures with separate support for each coefficient are different models because the partial derivatives cannot be matched The choice of space is thus a choice of where to place distributional assumptions and restrictions, not a choice between rival models In all other cases, either WTP must still be computed as a ratio of partial derivatives, or the WTP-space model is not equivalent to a preference-space model The paper concludes that analysts should report the implied distributions of both the partial derivatives of utility and the WTP measures References Balcombe, K., Chalak, A., and Fraser, I.
Ruth: (2009).
Warren: Model selection for the mixed logit with Bayesian estimation Berndt, E.
Ruth: K., Hall, B.
Paper summary: Ruth: So to recap this paper, "Willingness-to-pay space and preference space: a unified treatment of equivalence, heterogeneity and interpretation," the main thrust is bringing together scattered insights about these two modeling approaches.
Warren: It claims that willingness-to-pay space is actually just a reparameterization of utility where marginal utilities are written as the product of a cost coefficient and a monetary valuation instead of being in preference space.
Iris: The key claim they make is that if you have non-random coefficients, like when the cost coefficient is fixed, then the likelihoods from both spaces are identical after proper normalization.
Ruth: That’s important because it means for simpler models without random effects, we don't actually need to treat them as fundamentally different modeling frameworks; they just use different mathematical language to describe the same underlying choice structure.
Warren: But they spend a lot of time showing why this equivalence breaks down when you introduce deterministic heterogeneity in cost sensitivity or non-linear price effects. That’s where the differences really show up in terms of what we need to calculate.
Iris: The paper explains that with deterministic heterogeneity, you still have to compute willingness-to-pay as a ratio of partial derivatives, and this requires using those delta method standard errors if things get non-linear.
Ruth: That’s the practical takeaway: the space you choose dictates whether your estimates are directly estimable parameters or something you have to derive through calculus on top of the base preference space model.
Warren: And for random heterogeneity, they provide a specific condition for when latent class models maintain that equivalence, which is when each class has a non-random cost coefficient and the same allocation probabilities across those classes.
Iris: It’s about setting up clear rules so we know exactly when we can trust the equivalence claim and when we have to switch gears to use the more conservative ratio-of-derivatives approach.
Ruth: This paper matters because it clarifies that model choice isn't just about picking a "better" fit; it's fundamentally about where you place your assumptions regarding randomness and how you handle non-linearity.
Warren: It helps anyone who uses these spaces understand the subtle differences in what those parameters are actually doing economically, rather than just treating them as interchangeable labels.
Conclusion: Ruth: Looking at the title, "Willingness-to-pay space and preference space: a unified treatment of equivalence, heterogeneity and interpretation," it really tells you they are trying to solve a problem of misinterpretation.
Warren: Exactly. They aren't just saying WTP space is a different model; they are arguing that it’s a specific mathematical transformation, not necessarily an alternative theory for choice behavior.
Iris: The implication is that if you stick strictly to the formal conditions laid out in this paper, you can treat these two spaces as interchangeable tools for analysis.
Ruth: But the caveat is that those conditions are very specific about what kind of heterogeneity—random or deterministic—you’re dealing with. If your data doesn't fit those constraints, you have to revert to a different approach entirely.
Warren: So, for the listener on the road or at home, it means that when you see a parameter in WTP space, they need to know if that parameter is just a direct measure of value or if it’s a derived quantity through some sort of scaling factor.
Iris: The real implication is that analysts should report the implied distributions for both the partial derivatives of utility and those willingness-to-pay measures to be fully transparent about what they are doing.
Ruth: It moves the goal from finding one perfect model to being explicit about the assumptions you’re making about how people's preferences vary across individuals.
Warren: It’s a reminder that sometimes the most important thing isn't getting a better fit, but understanding exactly what you are fitting and what it takes to get there.
Stephane Hess
Choice Modelling Centre & Institute for Transport Studies, University of Leeds
econ.EM
Submitted: 2026-11-05
Updated: 2026-11-05
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 69/100
The gist: A growing number of choice modelling studies estimate their models “in willingness-to-pay (WTP) space”…
Key concepts
- WTP Space
- A reparameterization of utility where marginal utilities of attributes are written as a product involving a cost coefficient and a monetary valuation. This changes the numeraire of utility, effectively treating WTP as the attribute itself rather than just its preference level.
- Preference Space
- The traditional space where utility is defined by the direct preferences people have for different attributes. In this space, parameters represent direct subjective valuations of goods or services without explicitly incorporating a cost structure into the utility function.
- Equivalence Conditions
- These are the mathematical requirements under which a preference-space and WTP-space model produce identical results. For deterministic heterogeneity, they require matching specific partial derivatives and constants. With random heterogeneity, equivalence depends on the joint distribution of these parameters across individuals.
- Non-linearity and Practical Limitations
- The WTP space convenience holds only when cost effects are linear in utility. When non-linear price effects exist, WTP must be calculated as a ratio of partial derivatives using standard errors from the preference space, indicating that the two spaces are not always interchangeable.
Terminology
Summary
A growing number of choice modelling studies estimate their models “in willingness-to-pay (WTP) space”, yet often treat this as a different model that fits better (or worse) than “preference space”, and misread what its parameters represent WTP space is a reparameterisation, and with non-random coefficients the two spaces give identical likelihoods, WTP estimates and standard errors
Willingness-to-pay space as a reparameterisation
The paper introduces WTP space as a reparameterisation of utility where the marginal utilities of non-cost attributes are written as the product of a cost coefficient and a monetary valuation, rather than “in preference space” This transformation involves changing the numeraire of utility rather than the units of an attribute For instance, writing Vjn = δj + βtc tcjn + βtcβv,tt tt jn + βtcβv,ch chjn (3) confirms that the WTP is positive for an attribute that reduces utility The paper explains how this reparameterisation is equivalent to a change of units if time entered the utility function in hours rather than minutes
Equivalence and equivalence conditions
Principle 1 states that a preference-space and a WTP-space specification with the same specification of the error term are the same model if and only if, up to the normalisations needed for identification, they imply for every person the same alternative specific constants and the same partial derivatives of utility with respect to every attribute, measured as in the data With random heterogeneity, the condition is that the joint distribution of these constants and partial derivatives across people is the same Proposition 1 confirms this equivalence by stating that if one model has maximum likelihood estimates obtained by estimating each model separately, then (i) the maximised log-likelihoods are identical and (ii) the asymptotic covariance matrix of ϕˆ obtained from WTP-space estimation equals the Delta method covariance J Var d(ˆθ) J
Deterministically heterogeneous models
When deterministic heterogeneity is present, the WTP measure is always defined as the ratio of partial derivatives in (2) The paper details how heterogeneity can enter in two places: through monetised utility via functions qtc,n or qk,n, or through the cost coefficient βtc,n For example, when income affects cost sensitivity with elasticity λρn = (incn/inc)λ, the correct WTP-space counterpart puts the interaction on cost within monetised utility as in equation (12)
Non-linearity and practical limitations
The convenience of WTP space, namely that the estimated parameters are the WTP measures, holds only when cost and all attributes enter monetised utility linearly With non-linear price effects, WTP must again be computed as a ratio of partial derivatives with Delta method standard errors, exactly as in preference space The paper suggests retaining a single directly estimated value at a reference point by centring covariates or normalising non-linear transformations at a reference point
Random heterogeneity and distribution choices
With random heterogeneity, the two spaces are no longer automatically equivalent, and differences in fit come from the distributions specified The equivalence of latent class models holds if each class has a non-random cost coefficient and the same class allocation probabilities However, discrete mixtures with separate support for each coefficient are different models because the partial derivatives cannot be matched
Conclusion on model choice
The choice of space is thus a choice of where to place distributional assumptions and restrictions, not a choice between rival models In all other cases, either WTP must still be computed as a ratio of partial derivatives, or the WTP-space model is not equivalent to a preference-space model The paper concludes that analysts should report the implied distributions of both the partial derivatives of utility and the WTP measures
References
Balcombe, K., Chalak, A., and Fraser, I. (2009). Model selection for the mixed logit with Bayesian estimation Berndt, E. p. 9 Free class probabilities πrm remove the first difference but not the second, and
only when each class is also given its own coefficients do both become the same unrestricted
latent class model, which, by Proposition 2, is the same in the two spaces. The same applies to
non-parametric approximations that use a large, fixed grid of support points, such as logit mixed
logit (Train, 2016), as a regular grid in one space does not map onto a regular grid in the other. , 2016; Chen, 2026). Models are described as
∗Choice Modelling Centre & Institute for Transport Studies, University of Leeds, s. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. the usual logsum measure of consumer surplus (Small and Rosen, 1981) no longer applies, and
compensating variation requires the more general approach of Dagsvik and Karlstr¨om (2005). 4 Implications
The convenience of WTP space, namely that the estimated parameters are the WTP measures,
holds only when cost and all attributes enter monetised utility linearly, with any deterministic
heterogeneity entering additively through the valuation parameters. the usual logsum measure of consumer surplus (Small and Rosen, 1981) no longer applies, and
compensating variation requires the more general approach of Dagsvik and Karlstr¨om (2005). 4 Implications
The convenience of WTP space, namely that the estimated parameters are the WTP measures,
holds only when cost and all attributes enter monetised utility linearly, with any deterministic
heterogeneity entering additively through the valuation parameters. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. 1 Latent class models
In a latent class model (e. Greene and Hensher, 2003), each support point is a class with its
own full coefficient vector θc = (βtc,c, βtt,c, βch,c), and the class allocation probabilities πnc can be
constant across people or depend on characteristics of person n, typically through a multinomial
logit class allocation model. 4 Models with random heterogeneity
With random heterogeneity the two spaces are no longer automatically equivalent, and this is
where the claims of “better fit” come from. Train and Weeks (2005), Hensher and Greene (2011),
Hole and Kolstad (2012) and Chen (2026) find that preference space fits better (although only
marginally for the best-fitting models in the latter two), while Scarpa et al. 4 Models with random heterogeneity
With random heterogeneity the two spaces are no longer automatically equivalent, and this is
where the claims of “better fit” come from. Train and Weeks (2005), Hensher and Greene (2011),
Hole and Kolstad (2012) and Chen (2026) find that preference space fits better (although only
marginally for the best-fitting models in the latter two), while Scarpa et al. 1 Latent class models
In a latent class model (e. Greene and Hensher, 2003), each support point is a class with its
own full coefficient vector θc = (βtc,c, βtt,c, βch,c), and the class allocation probabilities πnc can be
constant across people or depend on characteristics of person n, typically through a multinomial
logit class allocation model. Free class probabilities πrm remove the first difference but not the second, and
only when each class is also given its own coefficients do both become the same unrestricted
latent class model, which, by Proposition 2, is the same in the two spaces. The same applies to
non-parametric approximations that use a large, fixed grid of support points, such as logit mixed
logit (Train, 2016), as a regular grid in one space does not map onto a regular grid in the other. , 2010; Hess and Train, 2017, eq. The language
of a price coefficient “normalised to one” is, however, carried over to the usual implementation of
WTP space, where the error scale is fixed and the cost coefficient is estimated1
, not least when
WTP-space models are fitted with generalised multinomial logit (G-MNL) software such as gmnl
(Gu et al. n is
general: Ln(Ω) = EG
h QTn
t=1 PYnt (θ)
i
, (22)
continuous mixture: Ln(Ω) = Z QTn
t=1 PYnt (θ) f(θ | Ω) dθ, (23)
discrete mixture: Ln(Ω) = PC
c=1 πnc QTn
t=1 PYnt (θc), (24)
where f is the density of G and, in the discrete case, G places probability πnc on support
point θc, with the probabilities possibly depending on covariates of person n. Writing the
same model in WTP space means specifying a mixing distribution Gϕ
(ϕ | Ω
ϕ
) for ϕ, and since
PYnt (θ) = PYnt (h
−1
(ϕ)), a preference-space model with G and a WTP-space model with Gϕ are
the same model if and only if Gϕ
is the distribution of h(θ) under G. A class that ignores cost, as in models of cost non-attendance, can
be represented in preference space but not in WTP space, and a class with a cost coefficient
close to zero leads to very large valuations that are poorly identified in WTP space, even though
the model, and hence the implied WTP, is the same as in preference space. Any difference in
fit between “a latent class model in preference space” and “a latent class model in WTP space”
with the same number of classes and the same allocation model indicates convergence to different
local optima, which is common with latent class likelihoods (Hess and Bunch, 2026) and is made
more likely by the non-linearity of the WTP-space utility. Preference space thus has the advantage of
avoiding a utility specification that is non-linear in parameters. With mixtures, local optima are
a concern in both spaces, but the non-linearity of the WTP-space specification further amplifies
the problem (Section 4). Since realistic specifications almost always include some deterministic
heterogeneity or non-linearity, not least in cost, this convenience is limited in practice. What can be retained is a single directly estimated value at a reference point, by centring
covariates, as with mean income in (12), or by normalising non-linear transformations, as in
(20). The label is misleading for two reasons. First, when the constants are not
1This ambiguity does not arise in software where the analyst writes the utility function directly, such as Apollo
(Hess and Palma, 2019). 4 Models with random heterogeneity
With random heterogeneity the two spaces are no longer automatically equivalent, and this is
where the claims of “better fit” come from. Train and Weeks (2005), Hensher and Greene (2011),
Hole and Kolstad (2012) and Chen (2026) find that preference space fits better (although only
marginally for the best-fitting models in the latter two), while Scarpa et al. Free class probabilities πrm remove the first difference but not the second, and
only when each class is also given its own coefficients do both become the same unrestricted
latent class model, which, by Proposition 2, is the same in the two spaces. The same applies to
non-parametric approximations that use a large, fixed grid of support points, such as logit mixed
logit (Train, 2016), as a regular grid in one space does not map onto a regular grid in the other. 4 Models with random heterogeneity
With random heterogeneity the two spaces are no longer automatically equivalent, and this is
where the claims of “better fit” come from. Train and Weeks (2005), Hensher and Greene (2011),
Hole and Kolstad (2012) and Chen (2026) find that preference space fits better (although only
marginally for the best-fitting models in the latter two), while Scarpa et al. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. Specification (15) has
a preference-space counterpart, in which income scales all non-cost coefficients, and in either
space only (12), like (10), represents an income effect operating through the marginal utility
of income, as is also the case when covariates are interacted with each valuation parameter
separately (e. Ladenburg and Skotte, 2022). Specification (15) has
a preference-space counterpart, in which income scales all non-cost coefficients, and in either
space only (12), like (10), represents an income effect operating through the marginal utility
of income, as is also the case when covariates are interacted with each valuation parameter
separately (e. Ladenburg and Skotte, 2022). Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. Alternative specific
constants included in monetised utility are valuations like any other and are included in θ and
ϕ, while constants outside monetised utility, as δj in (1), are specified in the same way in both
spaces and can be left out. Heterogeneity is represented by a mixing distribution G(θ | Ω), where
Ω contains the parameters of the distribution, such as means and covariances, and we write
PYnt (θ) for the probability, conditional on θ, of Ynt, the alternative chosen by person n in choice
situation t = 1, . Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. Heterogeneity is represented by a mixing distribution G(θ | Ω), where
Ω contains the parameters of the distribution, such as means and covariances, and we write
PYnt (θ) for the probability, conditional on θ, of Ynt, the alternative chosen by person n in choice
situation t = 1, . We use the standard specification for repeated choices of Revelt and
Train (1998), in which the coefficients vary across people but are constant across the choices
made by the same person. Balcombe et al. ,
2009; Rose and Masiero, 2010; Owusu Coffie et al. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. , 2007; Scarpa et al. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. the usual logsum measure of consumer surplus (Small and Rosen, 1981) no longer applies, and
compensating variation requires the more general approach of Dagsvik and Karlstr¨om (2005). 4 Implications
The convenience of WTP space, namely that the estimated parameters are the WTP measures,
holds only when cost and all attributes enter monetised utility linearly, with any deterministic
heterogeneity entering additively through the valuation parameters. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. Willingness-to-pay space and preference space:
a unified treatment of equivalence, heterogeneity and interpretation
Stephane Hess∗
6 November 2026
Abstract
A growing number of choice modelling studies estimate their models “in willingness-to-pay
(WTP) space”, yet often treat this as a different model that fits better (or worse) than
“preference space”, and misread what its parameters represent. This note brings together
and makes precise insights that have mostly been stated only in scattered places, organising
them around a single condition for two specifications to be the same model. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. , 2007; Scarpa et al. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al. Free class probabilities πrm remove the first difference but not the second, and
only when each class is also given its own coefficients do both become the same unrestricted
latent class model, which, by Proposition 2, is the same in the two spaces. The same applies to
non-parametric approximations that use a large, fixed grid of support points, such as logit mixed
logit (Train, 2016), as a regular grid in one space does not map onto a regular grid in the other. With random
heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are
equivalent only when the implied distributions match, as with full-covariance lognormals, and
independence is not a neutral assumption. 1 Introduction
Since Train and Weeks (2005), a growing number of studies specify discrete choice models “in
WTP space”, where the marginal utilities of non-cost attributes are written as the product of a
cost coefficient and a monetary valuation, rather than “in preference space”, where each attribute
has its own marginal utility (see also Sonnier et al.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Unified Model Equivalence Check: The improved AI system can automatically determine if two choice model specifications are equivalent by applying Principle 1, checking if they imply
for every person the same alternative specific constants and the same partial derivatives of utility with respect to every attribute, measured as in the data.
-
Automated Parameter Interpretation: The system can distinguish between a cost coefficient being
constrained to one
versus beingthe scale parameter
by analyzing how heterogeneity is placed on it, specifically noting thatthe label is misleading for two reasons.
-
Identification of Model Limitations: The system can flag when the convenience of WTP space is limited, identifying cases where
WTP must again be computed as a ratio of partial derivatives,
particularly when non-linearity or deterministic heterogeneity exists. -
Detection of Random Heterogeneity Effects: The AI can identify that
independence is not a neutral assumption
by analyzing how assumed independence in preference space impliesopposite correlation structures in the other space.
-
Distributional Assumption Comparison: The system can compare the implied distributions from different spaces, noting that
WTP space also makes it convenient to test restrictions on the WTP distribution directly
and thatthe choice of space is thus a choice of where to place distributional assumptions and restrictions, not a choice between rival models.
-
Non-Linear Transformation Handling: For non-linear cost effects, the system can automatically apply normalization at a reference point, such as
centring covariates or normalising non-linear transformations at a reference point,
to retain a single directly estimated value.
Abstract
A growing number of choice modelling studies estimate their models "in willingness-to-pay (WTP) space", yet often treat this as a different model that fits better (or worse) than "preference space", and misread what its parameters represent. This note brings together and makes precise insights that have mostly been stated only in scattered places, organising them around a single condition for two specifications to be the same model. WTP space is a reparameterisation, and with non-random coefficients the two spaces give identical likelihoods, WTP estimates and standard errors. With deterministic heterogeneity in cost sensitivity, or non-linearity, WTP must still be computed as a ratio of partial derivatives, although a simple normalisation recovers a directly estimated WTP at a reference point. With random heterogeneity, latent class models are equivalent in the two spaces, while other mixtures are equivalent only when the implied distributions match, as with full-covariance lognormals, and independence is not a neutral assumption.
Sources
Related papers
- Stationarity with Occasionally Binding Constraints
- Backtesting High-Dimensional Expected Shortfall
- Higher-Order Neyman Orthogonality in Moment-Condition Models
- Online Bayesian Inference for a Dirichlet Process Mixture of Binary Logit Models
- Debiased Conditional Independence Tests for Causal Effects
- Sharp Testable Implications of Encouragement Designs